Skip to content

ADR-0045 — Session calendar and cooldown semantics v1

Status: ACCEPTED

Date: 2026-09-26

Context

Wave 4's StrategySpec (CORE_CONTRACTS.md §21) already names SessionPolicy as one of its possible components, without defining its semantics. Decision Gate DG-F.1 (SCOPE.md, atom G04) must resolve that semantics — session calendars, cooldown policy, and timezone/DST integrity — before G04 can unblock H01/H03 per CAPABILITY_DAG.md.

G04 sits downstream of G02 (policy composition) and F06 (availability), and must respect the platform's existing temporal authority:

  • ADR-0006: no decision may use information unavailable at decision_time.
  • CORE_CONTRACTS.md §33: feature_available_time <= decision_time <= order_submit_time <= fill_time, and every derived value must be able to explain which source events it consumed and when they became available.

Bybit BTCUSDT itself is a 24/7 perpetual market: there is no Bybit trading calendar, no weekend closure, no exchange holiday. Any "session" semantics here are therefore not Bybit's own calendar — they are a strategy-facing temporal reference to the trading sessions of major traditional markets (New York, London, Tokyo), whose liquidity/volatility rhythms are commonly used as a signal even for instruments that never close.

PressurePolicyDefinition v1 (ADR-0028) already establishes an accepted, frozen shape for "deterministic policy over explicit evidence, evaluated at one explicit instant, with restrictions as upper bounds only, never a silent permissive default." Cooldown policy adopts that same shape rather than inventing a new idiom.

Decision

1. Session reference markets

Define three reference markets for v1: New York (NYSE/NASDAQ), London (LSE), Tokyo (TSE). These are not Bybit's own calendar (Bybit/BTCUSDT has none); they are named, independently evaluable temporal references a strategy may consult. A future second venue/instrument does not require touching this ADR; a future second reference market is additive in the same shape.

Each reference market has:

  • a market-open calendar: the set of UTC calendar dates on which it is open at all, derived from that market's real, documented holiday/closure calendar (not a bare Monday–Friday approximation);
  • zero or more named phases per open day, each an intraday time-of-day window converted to UTC. A market with no established phase structure simply has one phase ("regular") or the phases that are real for it — phase structure is not forced to be symmetric across reference markets.

Verified real, standard-time (non-DST) hours, converted to UTC (verified 2026-09-26; sources below):

New York (NYSE/NASDAQ), standard EST = UTC-5
  pre-market   09:00–14:30 UTC
  regular      14:30–21:00 UTC
  post-market  21:00–01:00 UTC (next day)

London (LSE), standard GMT = UTC+0
  pre-trading      05:05–07:50 UTC
  pre-open auction 07:50–08:00 UTC
  regular (core)   08:00–16:30 UTC
  closing auction  16:30–16:35 UTC
  post-trading     16:40–17:15 UTC

Tokyo (TSE), JST = UTC+9 fixed, no DST ever
  morning    00:00–02:30 UTC
  (lunch break 02:30–03:30 UTC: no phase, market closed for the break)
  afternoon  03:30–06:30 UTC (current hours, effective since the
                              November 2024 30-minute extension)

Tokyo has no established pre-market/post-market retail trading convention (unlike New York); London's extended-hours structure is auction-based (pre-open/closing auctions), not continuous trading like New York's pre/post-market. This asymmetry is a real fact about these markets, not an omission to fix.

Sources: NYSE/NASDAQ hours (stockmarkethours.org, TD Direct Investing); London Stock Exchange hours (globalexchangeclock.com, LSE public trading schedule); Tokyo Stock Exchange hours (Japan Exchange Group, official trading rules, jpx.co.jp — including the November 2024 afternoon-session extension).

2. Historical accuracy scope (v1)

  • DST transitions (New York, London — Tokyo has none) and holiday / closure calendars (all three markets, including one-off special closures that cannot be derived from any recurring rule) are both modeled the same way: an explicit, versioned static date table — never computed from a live host timezone database, and never derived from a calendar rule at evaluation time. This keeps evaluation deterministic and independent of the host system clock/tz data, which H05's bitwise-reproducible replay requires.
  • The table must cover at least 2020-01-01 through 2035-12-31 (Bybit BTCUSDT historical data begins 2020-03-25); extending it further is a data change, not a design change, and does not require reopening this ADR.
  • Early closes (e.g. a market closing at a modified time on a specific date) are entries in this same table — an exception overriding that date's normal phase end-time — not a separate mechanism from full-day closures.
  • Exchange session-hour rule changes over history (e.g. TSE's November 2024 30-minute afternoon-session extension) are explicitly not modeled retroactively for v1: the current rule applies uniformly across all historical replay. This is a deliberate, accepted v1 simplification, and it is intentionally asymmetric with DST/holiday handling above — DST and holiday closures recur predictably and indefinitely into the future, so a table costs little and stays exact; a market changing its own session-hour rule is rare and arbitrary, and modeling its full history is not required for this vertical.
  • The literal table contents (exact per-year DST transition instants, exact holiday/closure/early-close dates for 2020–2035, per market) are not fixed by this ADR — see Excluded.

3. Session data model and evaluation

Evaluating one named reference session at an explicit UTC instant answers exactly two questions: is this market open at all on this UTC calendar date (per its market-open calendar), and if so, which phase (if any) contains this instant. A caller receives an explicit "not open" result on a closed date — never a silently assumed phase.

No pre-named "derived" or overlap sessions (e.g. a hypothetical london_ny_overlap) are defined. A strategy computes overlaps itself by evaluating multiple named sessions and composing the per-session results — this keeps the primitive minimal and lets a strategy define whatever combination of reference markets it actually needs.

At any instant, zero, one, or multiple reference sessions may be simultaneously open. This is expected, not an error state: New York pre-market materially overlaps London's regular session, and a roughly 2.5 hour window exists between Tokyo's afternoon close (06:30 UTC standard time) and New York's pre-market open (09:00 UTC standard time) during which no reference market is open at all.

4. Trading-day boundary

Since BTCUSDT itself never closes, "trading day rollover" is a reporting/ accounting convention, not an exchange event. The trading-day boundary is UTC midnight, consistently with the frequency-limit day boundary in §6.

5. Excluded from session-calendar scope

  • Bybit's own trading calendar (it has none for BTCUSDT perpetual; this ADR does not invent one).
  • A generic multi-venue/multi-exchange calendar framework; only the three named v1 reference markets are in scope. A second venue's real trading hours (A12) are deferred until a concrete need exists, per SCOPE.md.
  • The literal DST/holiday/early-close table row data for 2020–2035 (verified data-compilation work — see Consequences).

6. Cooldown policy

CooldownPolicyDefinition v1 follows PressurePolicyDefinition's accepted shape: an immutable, versioned policy, evaluated at one explicit UTC as_of decision instant against explicit, caller-supplied realized-trade-history evidence. The evaluator never reads a host clock and never queries trade history itself.

Three named triggers combine into one evaluation (mirroring PressureDecision's single combined state, not three independent state machines a caller must reconcile):

  • Post-loss cooldown: any single realized loss — a position closed (per H04) with negative realized PnL, settled at or before as_of — starts a cooldown lasting a configured duration.
  • Consecutive-loss cooldown: N consecutive realized losses start a (typically longer) cooldown lasting a separately configured duration. Any realized win resets the consecutive-loss streak counter to zero unconditionally.
  • Intra-day frequency limit: no more than a configured number of new position entries — actually opened positions, not merely emitted DecisionIntents that were never executed — may start within one UTC calendar day (§4's boundary).

Composition rule: when the post-loss and consecutive-loss cooldowns are both active, the effective cooldown-until instant is the maximum (later) of the two. They extend each other; the later trigger never silently replaces or shortens the earlier one.

Invariants:

  • Only realized, settled, closed-position outcomes may trigger the post-loss or consecutive-loss cooldown. An open position's unrealized PnL never does — per ADR-0006/§33, an unrealized mark is not a causally-settled economic fact available for a decision to consume as trade history.
  • A cooldown restricts new entries only. It never blocks exit/reduction of an existing position: Risk/Exit policies remain free to close a position at any time regardless of cooldown state.
  • Like PressureRestrictions, a CooldownDecision's restrictions are upper bounds only: an ELIGIBLE result never itself authorizes a trade, it only means the cooldown gate does not forbid one.
  • Missing, stale, future-dated or malformed required trade-history evidence produces an explicit unavailable decision (mirroring PressureDecisionUnavailable) — never a silent default to ELIGIBLE.

Excluded

This ADR does not define or authorize:

  • the literal DST transition / holiday / early-close table contents for 2020–2035 (verified data-compilation work, not a semantic decision — see Consequences);
  • any second venue's or second instrument's session calendar (A12 remains deferred per SCOPE.md);
  • H01/H02/H04 Order/Fill/Portfolio semantics themselves, beyond the read-only realized-loss evidence CooldownPolicy consumes from them;
  • DG-F.2's execution conflict / intra-bar / OCO / partial-fill semantics (H03), a separate design gate;
  • a generic multi-market calendar or policy-engine framework.

Consequences

  • Issue #142 (G03/G04 implementation slice) can implement SessionPolicy and CooldownPolicyDefinition/CooldownDecision per this design without further semantic decisions. It must additionally compile the actual DST/holiday/early-close table for New York, London and Tokyo covering 2020–2035, citing authoritative per-exchange sources, as verified data — not as a new design question.
  • G04 can be credited RESOLVED in docs/architecture/OPEN_DECISIONS.md and docs/product/CAPABILITY_DAG.md once this ADR is accepted, per the governance-boundary rule that only an explicitly designated governance issue may make that edit (#147, not this one).
  • H01 (which requires G04 per CAPABILITY_DAG.md) remains blocked until issue #142's implementation is credited complete, not merely once this ADR is accepted.
  • DG-F.2 (H03, issue #140) is independent of this ADR's content and may proceed in parallel or in sequence per the repository's one-slice-at-a-time discipline, not per any dependency this ADR introduces.